The Illusion of Choice

This study measures records, not plans. When it says two plans cannot be told apart, it means every structured benefit field Pryse has loaded from the schemes’ own 2026 material holds the same value for both. It does not mean the plans are the same. Most records are thin, and thin records look alike.

That thinness is partly the schemes’ published material and partly the depth of Pryse’s own transcription, and the two cannot be separated from inside this dataset. Nothing here is evidence that a scheme failed to disclose something.

Each plan is reduced to 12 features. No feature carries a plan’s price level: rand amounts enter only as a ratio to the same plan’s own annual contribution, so scaling a plan’s whole rate card leaves its vector unchanged. Distance is Gower’s coefficient; grouping is agglomerative clustering under complete linkage, which joins two groups only when every pair across them is within the threshold.

• Medical savings share (proportion): Annual medical savings allocated to the main member, as a share of that member’s annual contribution, capped at 35%.
• Savings account loaded (yes/no): Whether the record carries any annual medical savings amount at all.
• Insured day-to-day share (proportion): Annual insured day-to-day limit as a share of the main member’s annual contribution, capped at 100%.
• Insured day-to-day loaded (yes/no): Whether the record carries an insured day-to-day limit separate from savings.
• Day-to-day limit scales with household (yes/no): Whether the insured day-to-day limit is published as a ladder that rises with household size rather than one figure.
• Named hospital restriction (yes/no): Whether the record names a restricted hospital list. Absence is not proof of unrestricted access, only that no restriction is loaded.
• Network GP requirement (yes/no): Whether a network-GP requirement is loaded. As above, absence records no loaded requirement rather than open access.
• Chronic co-payment rate (proportion): The single supported chronic-medicine co-payment percentage, scaled against a 50% ceiling. Zero where no single figure is loaded.
• Chronic co-payment figure loaded (yes/no): Whether a chronic co-payment percentage is loaded at all. Paired with the rate above so that "not loaded" can never be mistaken for "zero percent".
• Largest procedure co-payment (proportion): The largest published procedure co-payment as a share of the main member’s annual contribution, capped at 100%.
• Income-banded contributions (yes/no): Whether contributions vary by household income rather than being a flat rate.
• Children charged (proportion): How many children the scheme charges for before the rest are free, counted to a ceiling of four. Included because a family of five is genuinely covered differently under a three-child cap.

1. This measures the loaded record, not the plan. Two plans in the same group are ones whose records agree on every structured axis Pryse carries. Where a record is thin, they may still differ materially in ways no loaded field captures.
2. Evidence depth is a property of Pryse’s transcription as much as of the schemes’ disclosure, and the two cannot be separated from inside this dataset. A deeper transcription would raise the floor and split groups apart. The study is not evidence that a scheme failed to publish something.
3. Product-level formularies are absent for every record (PROJECT_CONTEXT.md §6), so chronic-medicine cover, one of the largest real differences between plans, is not measured at all.
4. Hospital-network composition is loaded for four of the records, so regional and network differences are largely invisible to the vector.
5. Price level is excluded from the comparison by design. The contribution figures quoted beside each group are main-member monthly rates for 2026; income-banded plans are represented by their top, open-ended band.
6. Nothing here says a plan is good, bad, cheap for what it offers, or suitable for anyone. Pryse is not a licensed financial services provider and this study ranks nothing by suitability.